Process modelling, automation and robotisation

The rational discovery, study, design and optimization of chemicals and materials is central to scientific, technological, economic and societal progress. However, these efforts are often hindered by long and expensive research and development cycles based on trial and error. Advances in artificial intelligence, machine-learning algorithms and their applications offer significant acceleration by enabling predictions and simulations of complex material properties and processes in quantitative agreement with experiments.

Objectives

The Process Modelling, Automation and Robotization group develops data-driven methods to tackle challenges in materials science and adjoining fields. Bridging the gap between experiments and simulations, it pioneers the accelerated discovery, study, design and optimization of novel materials, chemicals and their processes through automated experimental platforms and advanced simulations. The group’s contributions range from fundamental research to the development of solutions for real-world applications.

Scope of expertise

The group develops accurate, computationally efficient machine-learning surrogate models for functions that are expensive to evaluate, such as the results of wet-lab experiments or electronic-structure calculations. This includes machine-learning interatomic potentials for accurate all-atom simulations at unprecedented time and length scales, as well as multi-objective surrogate-based (Bayesian) optimization for property prediction and the design of chemicals and materials. These methods will drive semi-autonomous laboratories for materials discovery, such as developing novel water-splitting catalysts, the analyzing and optimizing vapour deposition processes, and accurately predicting thermal transport in nano-electronic devices.

Machine-learning methods include:

• Linear methods (e.g. least-squares regression, principal component analysis)
• Kernel methods (e.g. kernel ridge regression, Gaussian processes, support vector machines, kernel principal component analysis)
• Artificial neural networks (e.g. feedforward networks, convolutional networks)
• Deep learning (e.g. message-passing networks, equivariant networks, diffusion models)
• Learning with derivatives (e.g. machine-learning interatomic potentials)
• Other methods (e.g. subgroup discovery, symbolic regression)

Atomistic systems include:

• Elements (e.g. tungsten, warm dense hydrogen)
• Crystals and solids (e.g. zirconia, borosilicate glasses)
• Surfaces and interfaces (e.g. titania, alumina)
• Polymers and composite materials (e.g. vitrimers, flax fibre biocomposites)
• Small organic molecules (in vacuum and solvents)
• Drugs, drug-like molecules, and biomolecules (e.g. truxillic acid derivatives, triazoles, archazolid A)

Properties and processes include:

• Phase transitions (e.g. in H under pressure)
• Heat transport (e.g. via the Green-Kubo formalism)
• Nuclear chemical shifts (e.g. in organic molecules)
• Catalytic reactions (e.g. CO2 reduction)
• Deposition processes (e.g. gold on alumina)
• Drug-target binding (e.g. PPARγ, Farnesoid X, cyclooxygenase-1 receptors)

Features include:

• Local atomic environments (e.g. Coulomb matrix, many-body tensor, many-body expansions, atomic cluster expansion)
• Molecular structure graphs (e.g. descriptors, fingerprints, pharmacophores, graph kernels)
• Experimental measurements (e.g. of acoustic emissions, from ellipsometry)

Examples of applications:

• Machine-learning interatomic potentials (MLIPs) accelerate molecular dynamics simulations by several orders of magnitude compared to the underlying ab initio reference calculations. This enables accurate simulations at unprecedented time and length scales. The group develops ultra-fast potentials [xrh23], the fastest MLIPs at this time, contributed to the validation and benchmarking of MLIPs [bjr24, pcmt25, ppmt2025], and applied MLIPs to study atomistic systems and their processes [tjrc25].
• In quantum mechanics, the group leader demonstrated for the first time that the accurate prediction of atomization energies across chemical compound space is achievable. [rtml12] This was made possible by representations [lgr22], first the Coulomb matrix, and later the many-body tensor representation [hr22]. Follow-up publications [hmtm13] include the Δ-learning approach [rdrl15] and the widely used QM9 dataset [rdrl14].
• In density functional theory, the group leader laid the foundation for machine-learning functionals by demonstrating that machine learning can approximate density functionals. [srmb12] He further developed this line of research in follow-up work [srmb13, srmb15, vsmb15, lsmb15] and it has since become a research area in its own right.
• In physical chemistry, the group leader used machine learning to optimize transition-state-theory dividing surfaces [phmh12]. He also predicted the acid dissociation constants (pKa values) of monoprotic compounds [rkt11, rkt10] with kernel regression and graph kernels [rs10] tailored to small organic molecules [rps07], as well as nuclear chemical shifts [rrl15] in organic molecules.
• In medicinal chemistry, he identified a novel agonist of the diabetes-related transcription factor PPARγ (peroxisome proliferator-activated receptor γ) using Gaussian process-based virtual screening and cellular reporter gene assays. [rsms10] This truxillic acid derivative was further explored in subsequent studies. [srss10, ssss10]